[SPARK-29811] Missing persist on oldDataset in ml.RandomForestRegressor.train()

• Type: Sub-task
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 2.4.3
• Component/s: ML
• Assignee: Unassigned
• Reporter: IcySanwitch
• Created: 09/Nov/19 09:48
• Updated: 10/Nov/19 19:20
• Resolved: 10/Nov/19 19:20

Description

The rdd oldDataset in ml.regression.RandomForestRegressor.train() needs to be persisted, because it used in two actions in RandomForest.run() and oldDataset.first().

override protected def train(
      dataset: Dataset[_]): RandomForestRegressionModel = instrumented { instr =>
    val categoricalFeatures: Map[Int, Int] =
      MetadataUtils.getCategoricalFeatures(dataset.schema($(featuresCol)))
    val oldDataset: RDD[LabeledPoint] = extractLabeledPoints(dataset) // Needs to persist
    val strategy =
      super.getOldStrategy(categoricalFeatures, numClasses = 0, OldAlgo.Regression, getOldImpurity)

    instr.logPipelineStage(this)
    instr.logDataset(dataset)
    instr.logParams(this, labelCol, featuresCol, predictionCol, impurity, numTrees,
      featureSubsetStrategy, maxDepth, maxBins, maxMemoryInMB, minInfoGain,
      minInstancesPerNode, seed, subsamplingRate, cacheNodeIds, checkpointInterval)
   // First use oldDataset
    val trees = RandomForest
      .run(oldDataset, strategy, getNumTrees, getFeatureSubsetStrategy, getSeed, Some(instr))
      .map(_.asInstanceOf[DecisionTreeRegressionModel])
   // Second use oldDataset
    val numFeatures = oldDataset.first().features.size
    instr.logNamedValue(Instrumentation.loggerTags.numFeatures, numFeatures)
    new RandomForestRegressionModel(uid, trees, numFeatures)
  }

The same situation exits in ml.classification.RandomForestClassifier.train.

This issue is reported by our tool CacheCheck, which is used to dynamically detecting persist()/unpersist() api misuses.

Issue Links

• duplicates: SPARK-29818 Missing persist on RDD